What the numbers actually say
Spending on AI is near-universal and returns mostly are not. MIT's NANDA initiative studied enterprise adoption across 150 leader interviews, 350 employee surveys, and 300 public deployments — and found that roughly 5% of AI pilots produce real revenue acceleration while the rest stall with little to no measurable impact on the bottom line.
The buyers noticed. A survey of 830 IT decision-makers this year found the top way companies judge AI has flipped from "productivity" to hard financial impact — direct revenue and margin nearly doubled as the primary success metric.
The market has moved from "does it save time?" to "did it show up in the P&L?" — and by that test, 19 out of 20 projects are failing.
The detail almost everyone will miss
The failures are not a model problem. MIT's researchers were explicit: the gap is a "learning gap" — generic tools work brilliantly for an individual but stall inside a company because they never adapt to the company's actual workflows.
Two findings make it concrete. Companies that bought and partnered for a specific job succeeded about 67% of the time, while those that built their own tool from scratch succeeded roughly a third as often. And more than half of AI budgets went to sales and marketing — while the biggest measured returns sat in the unglamorous back office.
The 5% didn't have better AI. They pointed ordinary AI at one painful workflow and wired it into how the work already runs.
Why this matters if you run a business
The gap between the subscription you pay for and the results you don't see is a wiring problem, not a horsepower problem. A more expensive model won't close it; a tool that touches your real data and your real steps will.
There's an upside hiding in that. If the dividing line were budget or model access, small operators would lose by default — but it's integration and discipline, both of which are cheap.
The one advantage the giants can buy — raw AI capability — is the one thing that turns out not to decide the outcome. The Federal Reserve's own monitoring now shows small businesses adopting AI faster than large firms for the first time, precisely because the cost of entry collapsed to a monthly subscription.
What to do about it
The research doubles as a checklist. Every winning move in the data fits on one page — and none of it requires a bigger budget:
- Pick one workflow with a dollar attached. Not "roll out AI" — one repeatable task where you can name the cost of doing it the old way.
- Aim at the back office, not the storefront. The measured ROI sat in invoice handling, support triage, and operations — not the marketing copy that gets the budget.
- Wire it into the tools you already run. The win is the integration, not another browser tab someone forgets to open.
- Let the people who do the work drive it. Adoption stuck when it lived in a central "AI lab" and moved when line managers owned it.
- Set the number before you start. Decide what result would justify the effort, then measure against it — so you know whether you're in the 5% or the 95%.
None of these is a technology decision — every one is an operating decision, which is exactly why the technology was never the thing standing between you and the return.